Ponugoti Sushma

About

Built and productionized an LLM/RAG question-answering service over technical documentation, focusing on retrieval quality (reranking + IR metrics), latency, and scaling. Experienced orchestrating end-to-end ETL/ML workflows with Airflow/Prefect/AWS Step Functions and improving reliability via parallelism, retries, and shadow testing. Also delivered an explainable healthcare risk-flagging classifier with a stakeholder-friendly dashboard for a non-technical program manager.

Experience

Education

Key Strengths

• Built and deployed production RAG-based LLM QA system over technical documents
• Improved retrieval relevance via chunking/metadata filters and reranking
• Validated reranking with IR metrics (Recall@K, MRR), pairwise comparisons, and shadow A/B testing
• Performance/latency optimization using batching and code optimization
• Scaled infrastructure for growing document volume (DB instance upgrades, load balancing)
• Strong orchestration experience (Airflow/Prefect/Step Functions) for multi-step ML pipelines
• Stabilized pipelines by parallelizing embedding jobs, adding retries with exponential backoff, and tracking partial progress via Airflow XCom
• Systematic agent/workflow evaluation with offline tests, synthetic edge cases, shadow mode, and continuous monitoring
• Effective collaboration with non-technical healthcare stakeholder; translated requirements into an explainable classifier + dashboard and iterated on thresholds using historical cases